All right, everyone. I am Brandon Odo, a critical care PA from Connecticut, and what I'd love to get in with you today is really just a couple things. Why we seem to have a hard time effectively diagnosing a lot of our patients in the ICU, and maybe a couple tricks for doing it better. We don't have time for more than a quick review, certainly don't have time for a whole lot of data, but the data that you will find is probably going to be in autopsies. There have been a number of ICU autopsy studies looking at ICU deaths and comparing their pre- and post-mortem diagnoses, and to summarize, regardless of the population you look at, we seem to be missing a fair amount, both missed and missed diagnoses. That being said, I'm not really sure what this can say about the patients who don't die, because really I think the most common error we might be making is neither missed or missed diagnosis, but just failure to fully unpack diagnoses. How many patients are in your ICU right now with labels like this, sepsis, we don't know why, respiratory failure? What disease is acute hypoxic respiratory failure? Can you show me that in a textbook? But a lot of patients never really get beyond these labels, and I don't know if that's a problem. You could say that if you treat them supportively and empirically, and they get better, maybe that's okay, but I don't know. First of all, some patients won't get better if you don't offer them a specific treatment, but even the ones who do, if you never identify what you're treating, the best you can do is overtreat it, which means that you're giving too much, and we all know that there are complications associated with that. Some diseases will occur again if you don't identify them. How can you meaningfully engage in shared decision making and prognostication if you don't know what you're treating? Doc, what are my chances? I don't know, I don't know what you have. There are public health implications to some diagnoses, like toxicologic exposures, communicable diseases that could affect other people. Imagine being the first person to see COVID-19 and waving it away as respiratory failure. And I think there's just something toxic to what it does to us when we stop treating diseases specifically and mechanistically, and we get in the habit of treating symptoms supportively. The original diagnostician, Sherlock Holmes, who was, by the way, based on a physician, said there are just three things you need to make a diagnosis or solve a case, observation, knowledge, and deduction. Observation is the ability to look at the patient or the crime scene and glean the data from that. Knowledge is the other data, the background knowledge you have of how the world is, in our case, what diseases are like. And deduction is the ability to put all of that data on the table and connect the dots, see the pattern it's forming, and make a leap. So let's look at each of these steps and see where we're falling short. When it comes to observation, the most important tool you have is still the patient's history. They want to tell you what's wrong with them. But it is hard, right? Our patients are often confused, comatose, intubated, sick. So what do we do instead of good histories? We try to do these broad tests. We scan everything. We send chemistries and cultures. But they just don't take the place of it, do they? If you've ever seen this show House, you've seen this team of crack diagnosticians spend an hour doing bizarre tests to solve a medical mystery. A lot of the time could have been solved in the first couple minutes if you just asked the patient a couple questions. It's also true that our patients are sick, and often we have to prioritize treating them instead of diagnosing them. And sometimes treating helps. A trial of therapy can tell us something useful. But if it doesn't answer the question, it could also muddy the waters. Knowledge is why diagnosis is a lifelong journey. We just can't take the place of a really deep and broad base of knowledge of what diseases are like. And if you don't know what a disease looks like, you're never going to make that diagnosis. And yet, what do we focus on in critical care? When you walk around at this conference, the diagnostic focus, if there is one, is going to be on our ICU problems, ARDS, sepsis, delirium, most of which really are just syndromes that occur in the setting of other diseases. And those diseases are not our focus. They're another specialty's problem. But of course, the patients don't know what chapter they're in. Finally, deduction. Well, it probably never seems like the right moment to sit down and sink your teeth into a medical mystery. Right? We're all busy. But I think that's often an excuse more than anything. We just don't try a lot of the time. We're happy with these fake diagnoses and empiric treatments. And what this really ends up looking like is data collection is not a snapshot. You look at what you have. It generates a hypothesis, and that suggests the next test you need. And if we ever do this, we ever try, what it looks like is standing around on rounds, stroking our chins. We have an idea and say, could it be X? And so we check another box or two. We send a test. But a lot of the time, what it should suggest is asking another question. But do we do that? How often do you go back for more history after the patient's admitted? I think even in the Star Trek future, when we can run a tricorder down the patient, scan every cell in their body, we're still going to need to ask these questions. How are you feeling? If you had weight loss, how's your breathing? So in the real world, what do we do? We try to test everything. And it doesn't work, not because it's costly or inefficient, which it is, but it just doesn't work. There's too many things to test for if you can't get into the right ballpark first. We try to consult our way out, which doesn't work either. The consultants know additional things, but they don't have the continuity we have. They probably don't care as much. They have a lot of other patients. So if you have a specific question for them, they could be helpful, but if your question is just what's going on, they probably don't have much to say. We say somebody else will diagnose this patient on the floors or in the outpatient setting. How often do you think that happens? These people have their own problems, they have a lot more patients, and now the problem has aged. And we try to do it the same way every time. And just as you can't treat every patient in your busy unit like a medical mystery, I also don't think you can treat the ones that seem a little off as if they are a bread and butter case. You need the ability to switch modes a little bit and slow down and be more systematic. So you kind of have to start over in many cases. Use the data you have, but discard the differential you have and look at things with fresh eyes. Kind of admit the patient new and go from there. You know, the hypotheses you've had so far have gotten you where you are, which is not good enough. And I think a practical way to do this a lot of the time, in, again, the busy worlds we live in, is to make day two of ICU the day you diagnose. It's not practical to think that you're going to diagnose everyone fully when they're admitted. They're sick, they're being resuscitated, you don't have all the data back yet, and oftentimes it may be after hours, skeleton staffing. But day two, the dust has settled, you're in the light of day, you can look at what you found, decide what you still don't understand, make a fresh differential, and then as part of that process, expect to go and get more history. Don't get in that cycle of just trying to send a test. Plan to go back and ask some more questions. And if you really make that a habit, I think you'll make a lot more diagnoses than you are now. It's kind of like the tertiary survey the surgeons do. They know they're going to miss things initially because so much is going on, so they plan for it. As you go through all this, you're going to expect that you're going to be very fallible as we all are. So rather than trying to remember everything, you're going to write things down, the abnormalities, your differential. You're going to try to label what you're looking at as specifically as you can, so you can trigger those pathways and buzzwords you've learned about because that's how you're going to engage your pattern recognition. You're going to really reflect on the test characteristics of those tests you use, so you don't falsely rule something in or out when that test didn't really have the power to do that. And that can be subtle and may involve looking things up, talking to experts. And you're going to use whatever tools you can to back you up, because this is hard. And just like the airline pilot's not going to assume he remembers every step of his preflight checklist, you're going to assume that you might miss things. You probably have cognitive biases you're bringing to the table. Now it's not particularly clear from the data whether acknowledging these helps. You can't make them go away. I think that you can maybe take the edge off if you acknowledge them. The common ones are going to be anchoring, sticking with one diagnosis longer than you ought to. Availability or recency bias. Every diagnosis you hear about at this conference, you're going to try to go make for the next week or month or year. And if you have an interest in diagnosis, which perhaps you do if you're in this room, you're probably always going to be a little bit drawn towards the good diagnosis. The fun one, the rare one, the elegant explanation. Even if it's not the most likely, just based on prevalence. If you acknowledge that, then you can try to mitigate it a little bit. Here are some checklists. These QR codes will take you there. I put these together. They're not perfect. You're going to have to optimize them for you. I would suggest you either put it on your electronic device so you can pull it up or just dump it into your EMR as something like a dot phrase. That would be very easy to get. This first one is just questions. It's the elements of the history you might not always remember to make use of. You see the ones that are applicable and you say, oh, I should ask that. The second one is the same for physical exam maneuvers. And the last is diagnoses. The things you may not always remember that are perhaps high yield, morbid, specifically treatable and easily missed. You don't use these in place of your reasoning. You use them to supplement at the end of it. You get as far as you can and then you check your lists or your tools and you say, what might I have missed here? So where does this leave us? I think the first step to a lot of this is just making an effort to diagnose. And if you just do that, I think you'll get a lot farther than we are on a lot of our patients. And a lot of the next steps is just tools you already have. We all learned in school how to take a good history, do a good exam, build a differential. To some extent, I think we just forgot. We got into the real world where it's busy. We saw common things a lot and we focus, especially in critical care, on treatment over diagnosis. So you have the tools in your toolbox, it's just maybe a matter of using them. And finally, I hope you can let yourself be a little bit excited by all this. Diagnosis is still one of the most complex and I like to think the most interesting things we do in medicine. Perhaps the last thing we'll lose to the machines and really one that requires us much more than a lot of treatment, which can often be somewhat protocolized. It's my opinion, I'm sticking to it. This is me and I'm happy to answer any questions or if you missed a checklist or something just reach out. I'll give you over to Casey here.

ICU diagnosis

missed diagnoses

cognitive biases

patient history

diagnostic accuracy